Effective Image Database Search via Dimensionality Reduction

Anders Bjorholm Dahl and Henrik AanæsIEEE Computer Society Conference on

Computer Vision and Pattern Recognition Workshops

Outline

Introduction Methods

LF-clustering Experiments and Results Discussion and Conclusion

Introduction

The bag-of-words approach1. Feature extraction from the database

images2. Building the bag-of-words

representation3. Searching with a query image

Introduction

The Bag-of-word Model

Methods

Feature representation Clustering Feature assignment Image matching

Feature representation

PCA is applied to reduce the dimensionality of the feature vectors

The reduction of the SIFT descriptor is from 128 to between 3 and 12 dimensions

After dimension reduction we add color to our features the mean RGB value in a 10 × 10 pixels

patch around the localization of each feature

Feature representation

is the PCA reduced SIFT feature is the mean RGB values is a weighing parameter

( ) 1. normalized to unit length2. normalized

[ , (1 ) ]PCA RGBs s s

PCAs

RGBs

0.5 ,PCA RGBs s

s

Clustering

Similar but faster than Mean-shift clustering

Feature assignment

Similarity of images are found by comparing frequency vectors of a query image to images in the database

Give each visual words a weight[16]

log( )

: the weight of word

: the total number of images in the database

: the number of images where word occurs

ii

i

i

Nw

n

w i

N

n i

[16] D. Nister and H. Stewenius. Scalable recognition with a vocabulary tree. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 2, pages 2161–2168, June 2006.

Image matching

Frequency vectors are compared using the norm which is found to be superior to the

euclidean distance[16]

norm gives equal weight to the overlapping and non-overlapping parts

Inverted files are used for fast image retrieval

1L

1L

Experiments and Results

Data set first 1400 images form [16]

a series of 4 images of the same scene Use three of the images from one scene

to train the model and the last for testing

The test result is the percentage of the correct images ranked in top 3

data set is relatively smallhttp://www.vis.uky.edu/~stewe/ukbench/

Experiments and Results

Data set：

Experiments and Results

Experiments Color added PCA SIFT

3, 8, and 12 dimensional PCA SIFT featuresadded features are 6, 11, and 15 dimensions

compare with SIFT features reduced with PCA to 6, 11 and 15 dimensions (without color)

Clustering experiments LF-clustering

from 8,000 to 12,000 clusters k-means

10 clusters in 4 levels resulting in 10,000 clusters

Experiments and Results

Results

Experiments and Results

Results

Discussion and Conclusion

did not apply LF-clustering to the 128 dimensional SIFT features, because it performed very poorly

for future work the model should be tested on a larger set of data

A problem of the design of the bag-of-words model is it static nature not designed for adding or removing

images from the database